324 XI ELEMENTARY SPECTRAL THEORY
(b) Each number A ^ 0 in the spectrum is an eigenvalue of u.
(c) For each A^Q in S, there is a unique decomposition of E into a
topological direct sum of two subspaces F(X),
N(A) (also written F(l; u),
N(li uj) such that:

(i) F(/l) is closed, N(A) is finite dimensional;
(ii) «(F(A)) c F(l), awJ /Ac restriction of u - A • 1E /o F(A) w a ///tear
homeomorphism of that space onto itself,

(iii) w(N(A)) c N(A) #/?£/ /Aere is a smallest integer k = k(X), called the
order ofl (also written k(l;
«)), such that the restriction to N(/l) of(u — l- lE)fc
isO.

(d) The eigenspace E(A) 0/ w corresponding to the eigenvalue X ^ 0 w
contained in N(A) (hence finite dimensional).

(e) IfX, fi are two different points 0/S, distinct from 0, /Ae/z N(/j) c F(/l).
(f) ^E w a Banach space, the function ( -* (u - C " IE)" *» wA/cA w defined
and analytic in
C — S, Aoy a /?0/£ of order k(l) at each point A ^ 0 o/S.

Let /I ^ 0 be any complex number; as I'1 u is compact, we can apply the
Riesz theory (Section 11.3). By (11.3.4), if A is not an eigenvalue of u, 1EA~lu
is a linear homeomorphism of E onto itself, and the same is true of course
of M —A • 1E= —A(1E — A""1^), i.e. A is regular for w, which proves b).
Suppose on the contrary A is an eigenvalue of u; then the existence of the
decomposition F(X) 4- N(A) of E with properties (i), (ii), (iii), follows from
(11.3.3), as well as (d) (E(X) is the kernel noted Nx in (11.3.3)). To end the
proof of (c), we need only show the uniqueness of F(/l) and N(/l). Suppose
there is a second decomposition E = F' + N' having the same properties, and
write v = u — 1-1E. Then, any xeN' can be written x = y + z where
y e F(/l), z e N(A); by assumption there is h> 0 such that vh(x) = 0, hence
vh(y) = 0; as the restriction to F(A) of vh is a homeomorphism by assumption,
7 = 0 and x e N(A). This proves that N' c N(A), and a similar argument
proves N(/l) c N'. Next, if .x = y + z e F' with y e F(/i), z e N(/l), we have
ifyc) = v\y), hence ^(F7) c F(A); but as i?(F) = F', this implies F' c F(X)9

Denote by ul9 u2 the restrictions of u to F(A) and N(A), respectively. From
the relation (w2 — ^' lN(A))/t=:0> it follows by linear algebra (A.6.10 and
A.6.12) that there is a basis of N(A) such that the matrix of u2A • 1N(A)
with respect to that basis is triangular with diagonal 0; if d- dim (N(A)),
the determinant of u2 — C * IN<A) is therefore equal to (A — C)d and this proves
that «2 ~ C ' IN<A> is invertible if C T^ A. Let us prove on the other hand that
ui - C * IF(A) is invertible for C - A small enough: we can write Wj - f • 1F(A) =
iY + (A — 0 • 1F(A) with v1 = z/j — A • 1F(A). We know by (c) that t^ is inver-
tible; by (5.7.4), we therefore have \\v^l(x)\\ ^ \\v^\\ \\x\\ in F(A), which can
also be written K(;>c)|| > c - \\x\\ with c=||t?r1||"1. Now if C ^ 0 and